[openib-general] MAD queuing model

[Hal Rosenstock]

On Wed, 2004-09-01 at 14:39, Sean Hefty wrote:
> > > Along this same line, should a MAD requiring 
> > > RMPP post multiple work requests or post a single request at a time,
> > > until it completes?  (By completion, I mean the work request only, 
> > > and not an RMPP response.)
> 
> I asked this more based on what the response to the first question was, 
> along with how the layering worked.  It also comes down to some fairness, 
> since while a large RMPP request is being sent, responses for other MADs 
> may be queued behind it, which could result in timeouts on other MADs.  
> Ideally, correct RMPP windowing would avoid this type of condition.

This is mainly an issue with the SA (responses which can transfer a lot
of data). (There may be other proprietary uses of RMPP too).

The RMPP window could help with this but is this sufficient for
achieving fairness ? If it is, great.

Is fairness primarily an issue once sends are being deferred ? It seems
that in order to be fair some progress should be made on non RMPP sends
and perhaps this can be a simple as 1 non RMPP send to every n RMPP
sends where n is a compile time parameter ? Other ideas ?

> > This brings another question to mind. Should timeout_ms be ignored for
> > send methods if supplied in the send WR ?
> 
> I didn't quite follow this question.

The previous comment "if deferred sending were to be done,
request/response timeouts (ib_mad_send_wr.timeout_ms) should not start
until the MAD is actually posted." brought to mind the meaning of 
ib_mad_send_wr.timeout_ms in light of the discussion on solicited and
unsolicted messages. The header file says the following:
Timeout value, in milliseconds, to wait for a response message.  Set to
0 if no response is expected.

I was trying to ask if the MAD method is Send and timeout_ms is not 0,
whether this timeout should be honored or ignored by the access layer.

-- Hal